Polyp image segmentation method of multilevel attention network based on boundary guidance

Through a multi-level attention network based on boundary guidance, combined with multi-scale features and boundary perception mechanism, the problem of poor boundary information processing in polyp image segmentation is solved, and more refined polyp boundary segmentation and higher segmentation accuracy are achieved.

CN118941587AActive Publication Date: 2024-11-12JIANGXI NORMAL UNIV
9 Cites 11 Cited by

Patent Information

Application Number
CN202411414098.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-11
Publication Date
2024-11-12
Estimated Expiration
2044-10-11

AI Technical Summary

Technical Problem

When the existing polyp image segmentation method deals with polyps with large differences in shape, size and texture, it is difficult to accurately segment, and neglecting boundary information leads to missed detection or inaccurate segmentation boundary, affecting the overall segmentation effect.

Method used

A multi-level attention network based on boundary guidance was designed, combining the pyramid visual Transformer-PVT-v2 backbone, multi-scale feature extraction module, parallel partial decoder, boundary perception module and boundary guidance-based multi-level attention module, through Multilevel features and boundary perception mechanisms perform feature aggregation and attention enhancement to generate fine polyp boundary and segmentation features.

Benefits of technology

Effectively aggregating low-level and advanced features improves the segmentation accuracy and segmentation effect of polyp boundaries, reduces boundary blur, and improves the learning ability and generalization performance of the model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a polyp image segmentation method of a multilevel attention network based on boundary guidance. The method comprises the following steps: S1, constructing a data set; s2, constructing a multi-level attention network based on boundary guidance, and obtaining multi-level features of the image; s3, a multi-scale feature extraction module obtains multi-scale features of the compression channel; s4, acquiring a global feature map # imgabs0 # in the parallel partial decoder; s5, acquiring a boundary sensing feature # imgabs1 # through a boundary sensing module; s6, obtaining a segmentation feature map # imgabs2 #, a segmentation feature map # imgabs3 # and a segmentation feature map # imgabs4 # of different areas of the concerned polyp image; and S7, constructing a loss function, minimizing the loss function to optimize parameters of the multi-level attention network based on boundary guidance, and generating a clear polyp boundary. According to the method, the low-level features and the high-level global features are effectively fused, the complementation of the low-level features and the high-level features is realized, and the influence of the fuzzy polyp boundary on the model is effectively solved.
Need to check novelty before this filing date? Find Prior Art